Modelling skill retention in Space Fortress using machine learning

Abstract

Research on the optimal training frequency for highly skilled professionals is not well established. Finding an optimal training frequency could presumably lower costs, maintain a higher performance and create a more pleasant work environment. Royal Netherlands Aerospace Centre (NLR) started doing research on skill retention/decay in highly skilled professionals such as fighter pilots. By collecting participant data from their version of the online game Space Fortress (SF) a retention model will be created. Ideally, the final model can be extrapolated to predict an optimal training schedule for pilots and other professionals. In this thesis suitable techniques to create an accurate forecasting model for SF are explored, by studying machine learning techniques applied in Time Series Forecasting (TSF) and Knowledge Tracing (KT). After reviewing the literature, the most promising techniques will be discussed. A recommendation regarding many aspects of the challenge will be given, with the main focus on interpolation and prediction using a Long Short-Term Memory (LSTM) in combination with feature engineering.